Abstract Background This study aimed to investigate the reproducibility of the magnetic resonance tumour regression grade (mrTRG) and its agreement with the pathological regression grade (pTRG) in patients with locally recurrent rectal cancer (LRRC). Methods All LRRC patients who underwent a resection between 2010-2018 after treatment with induction chemotherapy and neoadjuvant chemo(re)irradiation in whom a restaging MRI was available were retrospectively selected. All MRI scans were reassessed by two independent radiologists using the mrTRG, and the pTRG was reassessed by an independent pathologist. The interobserver agreement between the radiologists aswell as between the radiologists and the pathologist was assessed using theweighted kappa test. A sub-analysis was performed to evaluate the influence of the interval between imaging and surgery. Results Out of 313patientswithLRRCtreatedduring the studyperiod, 124patientswere selected. Interobserver agreement between the radiologists was fair (k = 0.28) using a two-tier grading system (i.e. mrTRG 1-2 versus mrTRG 3-5). For the lead radiologist, agreement with pTRG was moderate (k = 0.52; 95 percent CI 0.36-0.68) when comparing good (mrTRG 1-2, Mandard 1-2) and intermediate/poor responders (mrTRG 3-5, Mandard 3-5), and the agreement was fair between radiologist 2 and pTRG (k = 0.39; 95 percent CI 0.22-0.56). A shorter interval (< 7 weeks) between MRI and surgery resulted in an improved agreement (k = 0.69), compared with an interval > 7 weeks (k = 0.340). For the lead radiologist, the positive predictive value for predicting good responderswas 95 percent (95 percent CI 71 percent-99 percent), whereas this was 56 percent (95 percent CI 44 percent-66 percent) for the other radiologist. Conclusion This study showed that, in LRRC, the reproducibility of mrTRG amongst radiologists is limited and the agreement of mrTRG with pTRG is low. However, a shorter interval between MRI and surgery seems to improve this agreement and, if assessed by a dedicated radiologist, mrTRG may be a safe tool to predict good responders.
RkJQdWJsaXNoZXIy MTk4NDMw